Yikes, they have a submission leaderboard. As if the expected signal to noise ratio of a system like this wasn't low enough.
Next, they could shoot themselves in the other foot by offering prizes based on leaderboard placement. /s
AI Incident Database
61–70 of 115 posts
Re: AI Incident Database
#62Earlier quoted context omitted.
Their minimum standard for inclusion appears to be any collection of if-statements.
This is the same standard for any Startup claiming to leverage AI, so I think that's fair
This measure of “fairness” merely impairs the utility of the data. Any standard based on puff-piece marketing is inherently garbage. There’s no research or policy value in dragging our thinking down to the same level.
Unsurprisingly, many of their taxonomies are 30-50% labelled “Other”, a common warning flag of either junk data or lazy classifications.
Re: AI Incident Database
#63Bit of a stretch to label an organisations shitty website security an AI incident just because it's a portal to the AI chatbot. I guess aviation disasters should now include passengers privacy incidents from hacked booking websites?
(I am from the org behind the Incident Database) I agree about it being a stretch, but we err towards indexing boundary cases. This one turns on whether you would consider the web application as being part of ChatGPT or not. In aviation, there are many systems that are part of safely flying from point to point that are not part of the plane itself. The control tower, the runway markings, and the processes built aroun…
Just make it easy to filter out. Possibly make it the default to filter out.
Re: AI Incident Database
#64This is interesting, but the presence of so many incidents that ultimately have nothing to do with AI is disappointing. Consider this incident: https://incidentdatabase.ai/cite/73/#r1325 >Description: Through a crowdsourcing social media campaign in 2016, several journalists and researchers demonstrated that augmented reality locations in the popular smartphone game Pokemon Go were more likely to be in white neighbor…
Based on that quote it sounds like the crowdsourcing was used to analyse the locations in order to show the bias. The claim is that AI was used to determine the locations on the first place. If I recall rightly the Pokémon Go locations were derived in part from an earlier (and far better) game called Ingress. Players could submit candidate locations, but it's possible that this was mostly ignored, and/or some form of…
It's possible, but the longer description makes it sound like the source of the issue was that the locations were crowd-sourced from a biased sample:
>The game creator Niantic Labs revealed that the Pokemon Go map was derived from a previous augmented reality game, Ingress. Ingress crowdsourced its map from users, who tended to be young, male, white, and English-speaking.
Similarly, from the attached article:
>"It turns out Niantic, which makes Pokémon Go, relied on a map from a previous augmented reality game called Ingress, which was crowd-sourced from its mostly male, tech-savvy players," she wrote in a blog post. "The result is a high concentration of PokéStops in commercial and downtown areas of some cities, while there are typically fewer PokéStops in non-white or residential areas, if there are any at all."
It could be there is some other AI-related problem for Pokemon Go hidden in one of the other articles (they all seem to be about unrelated things - one is about Jynx, one is about the experiences of a black person playing the game, another is speculating the Pokemon Go is a CIA operation...), but the incident report itself doesn't make any allegations of an AI-related issue.
Re: AI Incident Database
#65"Loom incident database" By the Ned Ludd organisation History repeats itself
I take it you think OSHA was a bad idea and we should go back to the level of worker safety that was seen at the height of the Industrial Revolution?
Yes, OSHA needs to be abolished, along with dozens of other federal agencies.
Re: AI Incident Database
#66Yikes, they have a submission leaderboard. As if the expected signal to noise ratio of a system like this wasn't low enough.
I work on this project. Every submission has to be approved by an editor before it appears on the site or gets counted toward leaderboard stats. We're currently more constrained by lack of submissions than by low-quality submissions – it's not uncommon that someone will contact us asking about an incident that should be in the database but isn't yet.
Re: AI Incident Database
#67Bit of a stretch to label an organisations shitty website security an AI incident just because it's a portal to the AI chatbot. I guess aviation disasters should now include passengers privacy incidents from hacked booking websites?
It's a bit hard to take any of the aggregate stats seriously when they bucket anything from "AI made a mistake and killed someone" to "AI-Generated-Text-Detection Tools Reported for High Error Rates" into the same "incident count".
Incident severity and scale varies wildly between incidents. We provide the taxonomy feature to support organizations like CSET adding color to incidents that support more descriptive stats.
Re: AI Incident Database
#68> https://incidentdatabase.ai/cite/516 What, again? So that's what the "Open" in OpenAI refers to: the user data and credit card info. Now that finally makes sense.
Credit card what?
Re: AI Incident Database
#69This is interesting, but the presence of so many incidents that ultimately have nothing to do with AI is disappointing. Consider this incident: https://incidentdatabase.ai/cite/73/#r1325 >Description: Through a crowdsourcing social media campaign in 2016, several journalists and researchers demonstrated that augmented reality locations in the popular smartphone game Pokemon Go were more likely to be in white neighbor…
> Alleged: Niantic Labs developed and deployed an AI system, which harmed non-white neighborhoods and communities of color.
This is also a dubious interpretation because it assumes more people being random visitors to a neighborhood is somehow a benefit. The cynical side of me also says that if the majority of visitors were to non-white neighborhoods that would've also been called a harm.
Re: AI Incident Database
#70ChatGPT bug was not an "AI" incident.